Outlier Detection via Iterative Temperature Cooling

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing outlier detection methods, such as those using soft-sort operations, can produce inaccurate outputs due to sensitivity to outliers, leading to incorrect classification and shifted inlier distribution estimates.

Innovation Solution

An information processing apparatus and method that calculates the probability of each data point being an outlier using a temperature parameter, which is gradually lowered towards zero, reducing the outlier's influence and ensuring accurate detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a temperature parameter is used to calculate outlier probabilities, then the detection accuracy is improved, but the computational complexity increases due to multiple iterative steps

Engineering Contradiction:

Improveoutlier detection accuracyVSAvoidcomputational complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies periodic action by iteratively updating the temperature parameter through multiple cooling steps. The temperature is systematically reduced in discrete stages (e.g., t, t/2, t/4, ..., until t < threshold), allowing the system to progressively refine outlier probability calculations. This periodic adjustment enables accurate detection while maintaining structured computational flow.

Inventive Principle:

Principle #19Periodic action

Solution Approach 2:

The temperature parameter is made dynamic rather than static. It starts at an initial value and continuously adapts by being halved in each iteration until it falls below a predefined threshold. This dynamic adjustment allows the system to balance between exploration (higher temperature) and exploitation (lower temperature), improving detection accuracy without requiring fixed complex computations.

Inventive Principle:

Principle #15Dynamics

2Reliability

If the temperature parameter is lowered gradually with multiple steps, then the influence of outliers is reduced, but the processing time increases

Engineering Contradiction:

Improveoutlier detection reliabilityVSAvoidprocessing time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The temperature parameter is reduced in periodic discrete steps (halving at each iteration) rather than continuously. This periodic cooling schedule allows the system to process data at manageable intervals, reducing outlier influence progressively while avoiding excessive processing time that would result from continuous fine-grained adjustments.

Inventive Principle:

Principle #19Periodic action

Solution Approach 2:

The temperature parameter acts as a temporary, disposable control variable that is systematically discarded (reduced to near-zero) after serving its purpose of guiding the outlier detection process. This approach allows rapid progression through detection stages without maintaining complex long-term computational structures.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

Data Source

PatentUS20230334297A1Information processing apparatus, information processing method, and computer readable medium

Publication Date: 2023.10.19 NEC CORP

AI summary

An object of the present disclosure is to provide an information processing apparatus, an information processing method, and a non-transitory computer readable medium capable of producing an accurate output to detect outlier(s). An information processing apparatus according to the present disclosure includes at least one memory configured to store instructions; and at least one processor configured to execute the instructions to: calculate each probability of each data point being an outlier by using a temperature parameter t, wherein t>0; lower the temperature parameter t towards 0 with a plural of step; and output the probability.